4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <linux/posix_acl.h>
36 #include <asm/uaccess.h>
40 /* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existent name.
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
88 * [10-Sep-98 Alan Modra] Another symlink change.
91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
108 * any extra contention...
111 /* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
118 static int do_getname(const char __user *filename, char *page)
121 unsigned long len = PATH_MAX;
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
130 retval = strncpy_from_user(page, filename, len);
134 return -ENAMETOOLONG;
140 static char *getname_flags(const char __user *filename, int flags, int *empty)
144 result = ERR_PTR(-ENOMEM);
147 int retval = do_getname(filename, tmp);
151 if (retval == -ENOENT && empty)
153 if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
155 result = ERR_PTR(retval);
159 audit_getname(result);
163 char *getname(const char __user * filename)
165 return getname_flags(filename, 0, 0);
168 #ifdef CONFIG_AUDITSYSCALL
169 void putname(const char *name)
171 if (unlikely(!audit_dummy_context()))
176 EXPORT_SYMBOL(putname);
179 static int check_acl(struct inode *inode, int mask)
181 #ifdef CONFIG_FS_POSIX_ACL
182 struct posix_acl *acl;
184 if (mask & MAY_NOT_BLOCK) {
185 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
188 /* no ->get_acl() calls in RCU mode... */
189 if (acl == ACL_NOT_CACHED)
191 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
194 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
197 * A filesystem can force a ACL callback by just never filling the
198 * ACL cache. But normally you'd fill the cache either at inode
199 * instantiation time, or on the first ->get_acl call.
201 * If the filesystem doesn't have a get_acl() function at all, we'll
202 * just create the negative cache entry.
204 if (acl == ACL_NOT_CACHED) {
205 if (inode->i_op->get_acl) {
206 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
210 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
216 int error = posix_acl_permission(inode, acl, mask);
217 posix_acl_release(acl);
226 * This does the basic permission checking
228 static int acl_permission_check(struct inode *inode, int mask)
230 unsigned int mode = inode->i_mode;
232 if (current_user_ns() != inode_userns(inode))
235 if (likely(current_fsuid() == inode->i_uid))
238 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
239 int error = check_acl(inode, mask);
240 if (error != -EAGAIN)
244 if (in_group_p(inode->i_gid))
250 * If the DACs are ok we don't need any capability check.
252 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
258 * generic_permission - check for access rights on a Posix-like filesystem
259 * @inode: inode to check access rights for
260 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
262 * Used to check for read/write/execute permissions on a file.
263 * We use "fsuid" for this, letting us set arbitrary permissions
264 * for filesystem access without changing the "normal" uids which
265 * are used for other things.
267 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
268 * request cannot be satisfied (eg. requires blocking or too much complexity).
269 * It would then be called again in ref-walk mode.
271 int generic_permission(struct inode *inode, int mask)
276 * Do the basic permission checks.
278 ret = acl_permission_check(inode, mask);
282 if (S_ISDIR(inode->i_mode)) {
283 /* DACs are overridable for directories */
284 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
286 if (!(mask & MAY_WRITE))
287 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
292 * Read/write DACs are always overridable.
293 * Executable DACs are overridable when there is
294 * at least one exec bit set.
296 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
297 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
301 * Searching includes executable on directories, else just read.
303 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
304 if (mask == MAY_READ)
305 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
312 * We _really_ want to just do "generic_permission()" without
313 * even looking at the inode->i_op values. So we keep a cache
314 * flag in inode->i_opflags, that says "this has not special
315 * permission function, use the fast case".
317 static inline int do_inode_permission(struct inode *inode, int mask)
319 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
320 if (likely(inode->i_op->permission))
321 return inode->i_op->permission(inode, mask);
323 /* This gets set once for the inode lifetime */
324 spin_lock(&inode->i_lock);
325 inode->i_opflags |= IOP_FASTPERM;
326 spin_unlock(&inode->i_lock);
328 return generic_permission(inode, mask);
332 * inode_permission - check for access rights to a given inode
333 * @inode: inode to check permission on
334 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
336 * Used to check for read/write/execute permissions on an inode.
337 * We use "fsuid" for this, letting us set arbitrary permissions
338 * for filesystem access without changing the "normal" uids which
339 * are used for other things.
341 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
343 int inode_permission(struct inode *inode, int mask)
347 if (unlikely(mask & MAY_WRITE)) {
348 umode_t mode = inode->i_mode;
351 * Nobody gets write access to a read-only fs.
353 if (IS_RDONLY(inode) &&
354 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
358 * Nobody gets write access to an immutable file.
360 if (IS_IMMUTABLE(inode))
364 retval = do_inode_permission(inode, mask);
368 retval = devcgroup_inode_permission(inode, mask);
372 return security_inode_permission(inode, mask);
376 * path_get - get a reference to a path
377 * @path: path to get the reference to
379 * Given a path increment the reference count to the dentry and the vfsmount.
381 void path_get(struct path *path)
386 EXPORT_SYMBOL(path_get);
389 * path_put - put a reference to a path
390 * @path: path to put the reference to
392 * Given a path decrement the reference count to the dentry and the vfsmount.
394 void path_put(struct path *path)
399 EXPORT_SYMBOL(path_put);
402 * Path walking has 2 modes, rcu-walk and ref-walk (see
403 * Documentation/filesystems/path-lookup.txt). In situations when we can't
404 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
405 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
406 * mode. Refcounts are grabbed at the last known good point before rcu-walk
407 * got stuck, so ref-walk may continue from there. If this is not successful
408 * (eg. a seqcount has changed), then failure is returned and it's up to caller
409 * to restart the path walk from the beginning in ref-walk mode.
413 * unlazy_walk - try to switch to ref-walk mode.
414 * @nd: nameidata pathwalk data
415 * @dentry: child of nd->path.dentry or NULL
416 * Returns: 0 on success, -ECHILD on failure
418 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
419 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
420 * @nd or NULL. Must be called from rcu-walk context.
422 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
424 struct fs_struct *fs = current->fs;
425 struct dentry *parent = nd->path.dentry;
428 BUG_ON(!(nd->flags & LOOKUP_RCU));
429 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
431 spin_lock(&fs->lock);
432 if (nd->root.mnt != fs->root.mnt ||
433 nd->root.dentry != fs->root.dentry)
436 spin_lock(&parent->d_lock);
438 if (!__d_rcu_to_refcount(parent, nd->seq))
440 BUG_ON(nd->inode != parent->d_inode);
442 if (dentry->d_parent != parent)
444 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
445 if (!__d_rcu_to_refcount(dentry, nd->seq))
448 * If the sequence check on the child dentry passed, then
449 * the child has not been removed from its parent. This
450 * means the parent dentry must be valid and able to take
451 * a reference at this point.
453 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
454 BUG_ON(!parent->d_count);
456 spin_unlock(&dentry->d_lock);
458 spin_unlock(&parent->d_lock);
461 spin_unlock(&fs->lock);
463 mntget(nd->path.mnt);
466 br_read_unlock(vfsmount_lock);
467 nd->flags &= ~LOOKUP_RCU;
471 spin_unlock(&dentry->d_lock);
473 spin_unlock(&parent->d_lock);
476 spin_unlock(&fs->lock);
481 * release_open_intent - free up open intent resources
482 * @nd: pointer to nameidata
484 void release_open_intent(struct nameidata *nd)
486 struct file *file = nd->intent.open.file;
488 if (file && !IS_ERR(file)) {
489 if (file->f_path.dentry == NULL)
496 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
498 return dentry->d_op->d_revalidate(dentry, nd);
502 * complete_walk - successful completion of path walk
503 * @nd: pointer nameidata
505 * If we had been in RCU mode, drop out of it and legitimize nd->path.
506 * Revalidate the final result, unless we'd already done that during
507 * the path walk or the filesystem doesn't ask for it. Return 0 on
508 * success, -error on failure. In case of failure caller does not
509 * need to drop nd->path.
511 static int complete_walk(struct nameidata *nd)
513 struct dentry *dentry = nd->path.dentry;
516 if (nd->flags & LOOKUP_RCU) {
517 nd->flags &= ~LOOKUP_RCU;
518 if (!(nd->flags & LOOKUP_ROOT))
520 spin_lock(&dentry->d_lock);
521 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
522 spin_unlock(&dentry->d_lock);
524 br_read_unlock(vfsmount_lock);
527 BUG_ON(nd->inode != dentry->d_inode);
528 spin_unlock(&dentry->d_lock);
529 mntget(nd->path.mnt);
531 br_read_unlock(vfsmount_lock);
534 if (likely(!(nd->flags & LOOKUP_JUMPED)))
537 if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
540 if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
543 /* Note: we do not d_invalidate() */
544 status = d_revalidate(dentry, nd);
555 static __always_inline void set_root(struct nameidata *nd)
558 get_fs_root(current->fs, &nd->root);
561 static int link_path_walk(const char *, struct nameidata *);
563 static __always_inline void set_root_rcu(struct nameidata *nd)
566 struct fs_struct *fs = current->fs;
570 seq = read_seqcount_begin(&fs->seq);
572 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
573 } while (read_seqcount_retry(&fs->seq, seq));
577 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
589 nd->flags |= LOOKUP_JUMPED;
591 nd->inode = nd->path.dentry->d_inode;
593 ret = link_path_walk(link, nd);
597 return PTR_ERR(link);
600 static void path_put_conditional(struct path *path, struct nameidata *nd)
603 if (path->mnt != nd->path.mnt)
607 static inline void path_to_nameidata(const struct path *path,
608 struct nameidata *nd)
610 if (!(nd->flags & LOOKUP_RCU)) {
611 dput(nd->path.dentry);
612 if (nd->path.mnt != path->mnt)
613 mntput(nd->path.mnt);
615 nd->path.mnt = path->mnt;
616 nd->path.dentry = path->dentry;
619 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
621 struct inode *inode = link->dentry->d_inode;
622 if (!IS_ERR(cookie) && inode->i_op->put_link)
623 inode->i_op->put_link(link->dentry, nd, cookie);
627 static __always_inline int
628 follow_link(struct path *link, struct nameidata *nd, void **p)
631 struct dentry *dentry = link->dentry;
633 BUG_ON(nd->flags & LOOKUP_RCU);
635 if (link->mnt == nd->path.mnt)
638 if (unlikely(current->total_link_count >= 40)) {
639 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
644 current->total_link_count++;
646 touch_atime(link->mnt, dentry);
647 nd_set_link(nd, NULL);
649 error = security_inode_follow_link(link->dentry, nd);
651 *p = ERR_PTR(error); /* no ->put_link(), please */
656 nd->last_type = LAST_BIND;
657 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
660 char *s = nd_get_link(nd);
663 error = __vfs_follow_link(nd, s);
664 else if (nd->last_type == LAST_BIND) {
665 nd->flags |= LOOKUP_JUMPED;
666 nd->inode = nd->path.dentry->d_inode;
667 if (nd->inode->i_op->follow_link) {
668 /* stepped on a _really_ weird one */
677 static int follow_up_rcu(struct path *path)
679 struct vfsmount *parent;
680 struct dentry *mountpoint;
682 parent = path->mnt->mnt_parent;
683 if (parent == path->mnt)
685 mountpoint = path->mnt->mnt_mountpoint;
686 path->dentry = mountpoint;
691 int follow_up(struct path *path)
693 struct vfsmount *parent;
694 struct dentry *mountpoint;
696 br_read_lock(vfsmount_lock);
697 parent = path->mnt->mnt_parent;
698 if (parent == path->mnt) {
699 br_read_unlock(vfsmount_lock);
703 mountpoint = dget(path->mnt->mnt_mountpoint);
704 br_read_unlock(vfsmount_lock);
706 path->dentry = mountpoint;
713 * Perform an automount
714 * - return -EISDIR to tell follow_managed() to stop and return the path we
717 static int follow_automount(struct path *path, unsigned flags,
720 struct vfsmount *mnt;
723 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
726 /* We don't want to mount if someone's just doing a stat -
727 * unless they're stat'ing a directory and appended a '/' to
730 * We do, however, want to mount if someone wants to open or
731 * create a file of any type under the mountpoint, wants to
732 * traverse through the mountpoint or wants to open the
733 * mounted directory. Also, autofs may mark negative dentries
734 * as being automount points. These will need the attentions
735 * of the daemon to instantiate them before they can be used.
737 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
738 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
739 path->dentry->d_inode)
742 current->total_link_count++;
743 if (current->total_link_count >= 40)
746 mnt = path->dentry->d_op->d_automount(path);
749 * The filesystem is allowed to return -EISDIR here to indicate
750 * it doesn't want to automount. For instance, autofs would do
751 * this so that its userspace daemon can mount on this dentry.
753 * However, we can only permit this if it's a terminal point in
754 * the path being looked up; if it wasn't then the remainder of
755 * the path is inaccessible and we should say so.
757 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
762 if (!mnt) /* mount collision */
766 /* lock_mount() may release path->mnt on error */
770 err = finish_automount(mnt, path);
774 /* Someone else made a mount here whilst we were busy */
779 path->dentry = dget(mnt->mnt_root);
788 * Handle a dentry that is managed in some way.
789 * - Flagged for transit management (autofs)
790 * - Flagged as mountpoint
791 * - Flagged as automount point
793 * This may only be called in refwalk mode.
795 * Serialization is taken care of in namespace.c
797 static int follow_managed(struct path *path, unsigned flags)
799 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
801 bool need_mntput = false;
804 /* Given that we're not holding a lock here, we retain the value in a
805 * local variable for each dentry as we look at it so that we don't see
806 * the components of that value change under us */
807 while (managed = ACCESS_ONCE(path->dentry->d_flags),
808 managed &= DCACHE_MANAGED_DENTRY,
809 unlikely(managed != 0)) {
810 /* Allow the filesystem to manage the transit without i_mutex
812 if (managed & DCACHE_MANAGE_TRANSIT) {
813 BUG_ON(!path->dentry->d_op);
814 BUG_ON(!path->dentry->d_op->d_manage);
815 ret = path->dentry->d_op->d_manage(path->dentry, false);
820 /* Transit to a mounted filesystem. */
821 if (managed & DCACHE_MOUNTED) {
822 struct vfsmount *mounted = lookup_mnt(path);
828 path->dentry = dget(mounted->mnt_root);
833 /* Something is mounted on this dentry in another
834 * namespace and/or whatever was mounted there in this
835 * namespace got unmounted before we managed to get the
839 /* Handle an automount point */
840 if (managed & DCACHE_NEED_AUTOMOUNT) {
841 ret = follow_automount(path, flags, &need_mntput);
847 /* We didn't change the current path point */
851 if (need_mntput && path->mnt == mnt)
855 return ret < 0 ? ret : need_mntput;
858 int follow_down_one(struct path *path)
860 struct vfsmount *mounted;
862 mounted = lookup_mnt(path);
867 path->dentry = dget(mounted->mnt_root);
873 static inline bool managed_dentry_might_block(struct dentry *dentry)
875 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
876 dentry->d_op->d_manage(dentry, true) < 0);
880 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
881 * we meet a managed dentry that would need blocking.
883 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
884 struct inode **inode)
887 struct vfsmount *mounted;
889 * Don't forget we might have a non-mountpoint managed dentry
890 * that wants to block transit.
892 if (unlikely(managed_dentry_might_block(path->dentry)))
895 if (!d_mountpoint(path->dentry))
898 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
902 path->dentry = mounted->mnt_root;
903 nd->flags |= LOOKUP_JUMPED;
904 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
906 * Update the inode too. We don't need to re-check the
907 * dentry sequence number here after this d_inode read,
908 * because a mount-point is always pinned.
910 *inode = path->dentry->d_inode;
915 static void follow_mount_rcu(struct nameidata *nd)
917 while (d_mountpoint(nd->path.dentry)) {
918 struct vfsmount *mounted;
919 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
922 nd->path.mnt = mounted;
923 nd->path.dentry = mounted->mnt_root;
924 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
928 static int follow_dotdot_rcu(struct nameidata *nd)
933 if (nd->path.dentry == nd->root.dentry &&
934 nd->path.mnt == nd->root.mnt) {
937 if (nd->path.dentry != nd->path.mnt->mnt_root) {
938 struct dentry *old = nd->path.dentry;
939 struct dentry *parent = old->d_parent;
942 seq = read_seqcount_begin(&parent->d_seq);
943 if (read_seqcount_retry(&old->d_seq, nd->seq))
945 nd->path.dentry = parent;
949 if (!follow_up_rcu(&nd->path))
951 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
953 follow_mount_rcu(nd);
954 nd->inode = nd->path.dentry->d_inode;
958 nd->flags &= ~LOOKUP_RCU;
959 if (!(nd->flags & LOOKUP_ROOT))
962 br_read_unlock(vfsmount_lock);
967 * Follow down to the covering mount currently visible to userspace. At each
968 * point, the filesystem owning that dentry may be queried as to whether the
969 * caller is permitted to proceed or not.
971 int follow_down(struct path *path)
976 while (managed = ACCESS_ONCE(path->dentry->d_flags),
977 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
978 /* Allow the filesystem to manage the transit without i_mutex
981 * We indicate to the filesystem if someone is trying to mount
982 * something here. This gives autofs the chance to deny anyone
983 * other than its daemon the right to mount on its
986 * The filesystem may sleep at this point.
988 if (managed & DCACHE_MANAGE_TRANSIT) {
989 BUG_ON(!path->dentry->d_op);
990 BUG_ON(!path->dentry->d_op->d_manage);
991 ret = path->dentry->d_op->d_manage(
992 path->dentry, false);
994 return ret == -EISDIR ? 0 : ret;
997 /* Transit to a mounted filesystem. */
998 if (managed & DCACHE_MOUNTED) {
999 struct vfsmount *mounted = lookup_mnt(path);
1004 path->mnt = mounted;
1005 path->dentry = dget(mounted->mnt_root);
1009 /* Don't handle automount points here */
1016 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1018 static void follow_mount(struct path *path)
1020 while (d_mountpoint(path->dentry)) {
1021 struct vfsmount *mounted = lookup_mnt(path);
1026 path->mnt = mounted;
1027 path->dentry = dget(mounted->mnt_root);
1031 static void follow_dotdot(struct nameidata *nd)
1036 struct dentry *old = nd->path.dentry;
1038 if (nd->path.dentry == nd->root.dentry &&
1039 nd->path.mnt == nd->root.mnt) {
1042 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1043 /* rare case of legitimate dget_parent()... */
1044 nd->path.dentry = dget_parent(nd->path.dentry);
1048 if (!follow_up(&nd->path))
1051 follow_mount(&nd->path);
1052 nd->inode = nd->path.dentry->d_inode;
1056 * Allocate a dentry with name and parent, and perform a parent
1057 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1058 * on error. parent->d_inode->i_mutex must be held. d_lookup must
1059 * have verified that no child exists while under i_mutex.
1061 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1062 struct qstr *name, struct nameidata *nd)
1064 struct inode *inode = parent->d_inode;
1065 struct dentry *dentry;
1068 /* Don't create child dentry for a dead directory. */
1069 if (unlikely(IS_DEADDIR(inode)))
1070 return ERR_PTR(-ENOENT);
1072 dentry = d_alloc(parent, name);
1073 if (unlikely(!dentry))
1074 return ERR_PTR(-ENOMEM);
1076 old = inode->i_op->lookup(inode, dentry, nd);
1077 if (unlikely(old)) {
1085 * We already have a dentry, but require a lookup to be performed on the parent
1086 * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1087 * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1088 * child exists while under i_mutex.
1090 static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1091 struct nameidata *nd)
1093 struct inode *inode = parent->d_inode;
1096 /* Don't create child dentry for a dead directory. */
1097 if (unlikely(IS_DEADDIR(inode))) {
1099 return ERR_PTR(-ENOENT);
1102 old = inode->i_op->lookup(inode, dentry, nd);
1103 if (unlikely(old)) {
1111 * It's more convoluted than I'd like it to be, but... it's still fairly
1112 * small and for now I'd prefer to have fast path as straight as possible.
1113 * It _is_ time-critical.
1115 static int do_lookup(struct nameidata *nd, struct qstr *name,
1116 struct path *path, struct inode **inode)
1118 struct vfsmount *mnt = nd->path.mnt;
1119 struct dentry *dentry, *parent = nd->path.dentry;
1125 * Rename seqlock is not required here because in the off chance
1126 * of a false negative due to a concurrent rename, we're going to
1127 * do the non-racy lookup, below.
1129 if (nd->flags & LOOKUP_RCU) {
1132 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1136 /* Memory barrier in read_seqcount_begin of child is enough */
1137 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1141 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1142 status = d_revalidate(dentry, nd);
1143 if (unlikely(status <= 0)) {
1144 if (status != -ECHILD)
1149 if (unlikely(d_need_lookup(dentry)))
1152 path->dentry = dentry;
1153 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1155 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1159 if (unlazy_walk(nd, dentry))
1162 dentry = __d_lookup(parent, name);
1165 if (dentry && unlikely(d_need_lookup(dentry))) {
1170 if (unlikely(!dentry)) {
1171 struct inode *dir = parent->d_inode;
1172 BUG_ON(nd->inode != dir);
1174 mutex_lock(&dir->i_mutex);
1175 dentry = d_lookup(parent, name);
1176 if (likely(!dentry)) {
1177 dentry = d_alloc_and_lookup(parent, name, nd);
1178 if (IS_ERR(dentry)) {
1179 mutex_unlock(&dir->i_mutex);
1180 return PTR_ERR(dentry);
1185 } else if (unlikely(d_need_lookup(dentry))) {
1186 dentry = d_inode_lookup(parent, dentry, nd);
1187 if (IS_ERR(dentry)) {
1188 mutex_unlock(&dir->i_mutex);
1189 return PTR_ERR(dentry);
1195 mutex_unlock(&dir->i_mutex);
1197 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1198 status = d_revalidate(dentry, nd);
1199 if (unlikely(status <= 0)) {
1204 if (!d_invalidate(dentry)) {
1213 path->dentry = dentry;
1214 err = follow_managed(path, nd->flags);
1215 if (unlikely(err < 0)) {
1216 path_put_conditional(path, nd);
1220 nd->flags |= LOOKUP_JUMPED;
1221 *inode = path->dentry->d_inode;
1225 static inline int may_lookup(struct nameidata *nd)
1227 if (nd->flags & LOOKUP_RCU) {
1228 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1231 if (unlazy_walk(nd, NULL))
1234 return inode_permission(nd->inode, MAY_EXEC);
1237 static inline int handle_dots(struct nameidata *nd, int type)
1239 if (type == LAST_DOTDOT) {
1240 if (nd->flags & LOOKUP_RCU) {
1241 if (follow_dotdot_rcu(nd))
1249 static void terminate_walk(struct nameidata *nd)
1251 if (!(nd->flags & LOOKUP_RCU)) {
1252 path_put(&nd->path);
1254 nd->flags &= ~LOOKUP_RCU;
1255 if (!(nd->flags & LOOKUP_ROOT))
1256 nd->root.mnt = NULL;
1258 br_read_unlock(vfsmount_lock);
1263 * Do we need to follow links? We _really_ want to be able
1264 * to do this check without having to look at inode->i_op,
1265 * so we keep a cache of "no, this doesn't need follow_link"
1266 * for the common case.
1268 static inline int should_follow_link(struct inode *inode, int follow)
1270 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1271 if (likely(inode->i_op->follow_link))
1274 /* This gets set once for the inode lifetime */
1275 spin_lock(&inode->i_lock);
1276 inode->i_opflags |= IOP_NOFOLLOW;
1277 spin_unlock(&inode->i_lock);
1282 static inline int walk_component(struct nameidata *nd, struct path *path,
1283 struct qstr *name, int type, int follow)
1285 struct inode *inode;
1288 * "." and ".." are special - ".." especially so because it has
1289 * to be able to know about the current root directory and
1290 * parent relationships.
1292 if (unlikely(type != LAST_NORM))
1293 return handle_dots(nd, type);
1294 err = do_lookup(nd, name, path, &inode);
1295 if (unlikely(err)) {
1300 path_to_nameidata(path, nd);
1304 if (should_follow_link(inode, follow)) {
1305 if (nd->flags & LOOKUP_RCU) {
1306 if (unlikely(unlazy_walk(nd, path->dentry))) {
1311 BUG_ON(inode != path->dentry->d_inode);
1314 path_to_nameidata(path, nd);
1320 * This limits recursive symlink follows to 8, while
1321 * limiting consecutive symlinks to 40.
1323 * Without that kind of total limit, nasty chains of consecutive
1324 * symlinks can cause almost arbitrarily long lookups.
1326 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1330 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1331 path_put_conditional(path, nd);
1332 path_put(&nd->path);
1335 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1338 current->link_count++;
1341 struct path link = *path;
1344 res = follow_link(&link, nd, &cookie);
1346 res = walk_component(nd, path, &nd->last,
1347 nd->last_type, LOOKUP_FOLLOW);
1348 put_link(nd, &link, cookie);
1351 current->link_count--;
1357 * We really don't want to look at inode->i_op->lookup
1358 * when we don't have to. So we keep a cache bit in
1359 * the inode ->i_opflags field that says "yes, we can
1360 * do lookup on this inode".
1362 static inline int can_lookup(struct inode *inode)
1364 if (likely(inode->i_opflags & IOP_LOOKUP))
1366 if (likely(!inode->i_op->lookup))
1369 /* We do this once for the lifetime of the inode */
1370 spin_lock(&inode->i_lock);
1371 inode->i_opflags |= IOP_LOOKUP;
1372 spin_unlock(&inode->i_lock);
1378 * This is the basic name resolution function, turning a pathname into
1379 * the final dentry. We expect 'base' to be positive and a directory.
1381 * Returns 0 and nd will have valid dentry and mnt on success.
1382 * Returns error and drops reference to input namei data on failure.
1384 static int link_path_walk(const char *name, struct nameidata *nd)
1394 /* At this point we know we have a real path component. */
1401 err = may_lookup(nd);
1406 c = *(const unsigned char *)name;
1408 hash = init_name_hash();
1411 hash = partial_name_hash(c, hash);
1412 c = *(const unsigned char *)name;
1413 } while (c && (c != '/'));
1414 this.len = name - (const char *) this.name;
1415 this.hash = end_name_hash(hash);
1418 if (this.name[0] == '.') switch (this.len) {
1420 if (this.name[1] == '.') {
1422 nd->flags |= LOOKUP_JUMPED;
1428 if (likely(type == LAST_NORM)) {
1429 struct dentry *parent = nd->path.dentry;
1430 nd->flags &= ~LOOKUP_JUMPED;
1431 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1432 err = parent->d_op->d_hash(parent, nd->inode,
1439 /* remove trailing slashes? */
1441 goto last_component;
1442 while (*++name == '/');
1444 goto last_component;
1446 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1451 err = nested_symlink(&next, nd);
1455 if (can_lookup(nd->inode))
1459 /* here ends the main loop */
1463 nd->last_type = type;
1470 static int path_init(int dfd, const char *name, unsigned int flags,
1471 struct nameidata *nd, struct file **fp)
1477 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1478 nd->flags = flags | LOOKUP_JUMPED;
1480 if (flags & LOOKUP_ROOT) {
1481 struct inode *inode = nd->root.dentry->d_inode;
1483 if (!inode->i_op->lookup)
1485 retval = inode_permission(inode, MAY_EXEC);
1489 nd->path = nd->root;
1491 if (flags & LOOKUP_RCU) {
1492 br_read_lock(vfsmount_lock);
1494 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1496 path_get(&nd->path);
1501 nd->root.mnt = NULL;
1504 if (flags & LOOKUP_RCU) {
1505 br_read_lock(vfsmount_lock);
1510 path_get(&nd->root);
1512 nd->path = nd->root;
1513 } else if (dfd == AT_FDCWD) {
1514 if (flags & LOOKUP_RCU) {
1515 struct fs_struct *fs = current->fs;
1518 br_read_lock(vfsmount_lock);
1522 seq = read_seqcount_begin(&fs->seq);
1524 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1525 } while (read_seqcount_retry(&fs->seq, seq));
1527 get_fs_pwd(current->fs, &nd->path);
1530 struct dentry *dentry;
1532 file = fget_raw_light(dfd, &fput_needed);
1537 dentry = file->f_path.dentry;
1541 if (!S_ISDIR(dentry->d_inode->i_mode))
1544 retval = inode_permission(dentry->d_inode, MAY_EXEC);
1549 nd->path = file->f_path;
1550 if (flags & LOOKUP_RCU) {
1553 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1554 br_read_lock(vfsmount_lock);
1557 path_get(&file->f_path);
1558 fput_light(file, fput_needed);
1562 nd->inode = nd->path.dentry->d_inode;
1566 fput_light(file, fput_needed);
1571 static inline int lookup_last(struct nameidata *nd, struct path *path)
1573 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1574 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1576 nd->flags &= ~LOOKUP_PARENT;
1577 return walk_component(nd, path, &nd->last, nd->last_type,
1578 nd->flags & LOOKUP_FOLLOW);
1581 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1582 static int path_lookupat(int dfd, const char *name,
1583 unsigned int flags, struct nameidata *nd)
1585 struct file *base = NULL;
1590 * Path walking is largely split up into 2 different synchronisation
1591 * schemes, rcu-walk and ref-walk (explained in
1592 * Documentation/filesystems/path-lookup.txt). These share much of the
1593 * path walk code, but some things particularly setup, cleanup, and
1594 * following mounts are sufficiently divergent that functions are
1595 * duplicated. Typically there is a function foo(), and its RCU
1596 * analogue, foo_rcu().
1598 * -ECHILD is the error number of choice (just to avoid clashes) that
1599 * is returned if some aspect of an rcu-walk fails. Such an error must
1600 * be handled by restarting a traditional ref-walk (which will always
1601 * be able to complete).
1603 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1608 current->total_link_count = 0;
1609 err = link_path_walk(name, nd);
1611 if (!err && !(flags & LOOKUP_PARENT)) {
1612 err = lookup_last(nd, &path);
1615 struct path link = path;
1616 nd->flags |= LOOKUP_PARENT;
1617 err = follow_link(&link, nd, &cookie);
1619 err = lookup_last(nd, &path);
1620 put_link(nd, &link, cookie);
1625 err = complete_walk(nd);
1627 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1628 if (!nd->inode->i_op->lookup) {
1629 path_put(&nd->path);
1637 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1638 path_put(&nd->root);
1639 nd->root.mnt = NULL;
1644 static int do_path_lookup(int dfd, const char *name,
1645 unsigned int flags, struct nameidata *nd)
1647 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1648 if (unlikely(retval == -ECHILD))
1649 retval = path_lookupat(dfd, name, flags, nd);
1650 if (unlikely(retval == -ESTALE))
1651 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1653 if (likely(!retval)) {
1654 if (unlikely(!audit_dummy_context())) {
1655 if (nd->path.dentry && nd->inode)
1656 audit_inode(name, nd->path.dentry);
1662 int kern_path_parent(const char *name, struct nameidata *nd)
1664 return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1667 int kern_path(const char *name, unsigned int flags, struct path *path)
1669 struct nameidata nd;
1670 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1677 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1678 * @dentry: pointer to dentry of the base directory
1679 * @mnt: pointer to vfs mount of the base directory
1680 * @name: pointer to file name
1681 * @flags: lookup flags
1682 * @path: pointer to struct path to fill
1684 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1685 const char *name, unsigned int flags,
1688 struct nameidata nd;
1690 nd.root.dentry = dentry;
1692 BUG_ON(flags & LOOKUP_PARENT);
1693 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1694 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1700 static struct dentry *__lookup_hash(struct qstr *name,
1701 struct dentry *base, struct nameidata *nd)
1703 struct inode *inode = base->d_inode;
1704 struct dentry *dentry;
1707 err = inode_permission(inode, MAY_EXEC);
1709 return ERR_PTR(err);
1712 * Don't bother with __d_lookup: callers are for creat as
1713 * well as unlink, so a lot of the time it would cost
1716 dentry = d_lookup(base, name);
1718 if (dentry && d_need_lookup(dentry)) {
1720 * __lookup_hash is called with the parent dir's i_mutex already
1721 * held, so we are good to go here.
1723 dentry = d_inode_lookup(base, dentry, nd);
1728 if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1729 int status = d_revalidate(dentry, nd);
1730 if (unlikely(status <= 0)) {
1732 * The dentry failed validation.
1733 * If d_revalidate returned 0 attempt to invalidate
1734 * the dentry otherwise d_revalidate is asking us
1735 * to return a fail status.
1739 return ERR_PTR(status);
1740 } else if (!d_invalidate(dentry)) {
1748 dentry = d_alloc_and_lookup(base, name, nd);
1754 * Restricted form of lookup. Doesn't follow links, single-component only,
1755 * needs parent already locked. Doesn't follow mounts.
1758 struct dentry *lookup_hash(struct nameidata *nd)
1760 return __lookup_hash(&nd->last, nd->path.dentry, nd);
1762 EXPORT_SYMBOL(lookup_hash);
1765 * lookup_one_len - filesystem helper to lookup single pathname component
1766 * @name: pathname component to lookup
1767 * @base: base directory to lookup from
1768 * @len: maximum length @len should be interpreted to
1770 * Note that this routine is purely a helper for filesystem usage and should
1771 * not be called by generic code. Also note that by using this function the
1772 * nameidata argument is passed to the filesystem methods and a filesystem
1773 * using this helper needs to be prepared for that.
1775 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1781 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1786 return ERR_PTR(-EACCES);
1788 hash = init_name_hash();
1790 c = *(const unsigned char *)name++;
1791 if (c == '/' || c == '\0')
1792 return ERR_PTR(-EACCES);
1793 hash = partial_name_hash(c, hash);
1795 this.hash = end_name_hash(hash);
1797 * See if the low-level filesystem might want
1798 * to use its own hash..
1800 if (base->d_flags & DCACHE_OP_HASH) {
1801 int err = base->d_op->d_hash(base, base->d_inode, &this);
1803 return ERR_PTR(err);
1806 return __lookup_hash(&this, base, NULL);
1809 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1810 struct path *path, int *empty)
1812 struct nameidata nd;
1813 char *tmp = getname_flags(name, flags, empty);
1814 int err = PTR_ERR(tmp);
1817 BUG_ON(flags & LOOKUP_PARENT);
1819 err = do_path_lookup(dfd, tmp, flags, &nd);
1827 int user_path_at(int dfd, const char __user *name, unsigned flags,
1830 return user_path_at_empty(dfd, name, flags, path, 0);
1833 static int user_path_parent(int dfd, const char __user *path,
1834 struct nameidata *nd, char **name)
1836 char *s = getname(path);
1842 error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1852 * It's inline, so penalty for filesystems that don't use sticky bit is
1855 static inline int check_sticky(struct inode *dir, struct inode *inode)
1857 uid_t fsuid = current_fsuid();
1859 if (!(dir->i_mode & S_ISVTX))
1861 if (current_user_ns() != inode_userns(inode))
1863 if (inode->i_uid == fsuid)
1865 if (dir->i_uid == fsuid)
1869 return !ns_capable(inode_userns(inode), CAP_FOWNER);
1873 * Check whether we can remove a link victim from directory dir, check
1874 * whether the type of victim is right.
1875 * 1. We can't do it if dir is read-only (done in permission())
1876 * 2. We should have write and exec permissions on dir
1877 * 3. We can't remove anything from append-only dir
1878 * 4. We can't do anything with immutable dir (done in permission())
1879 * 5. If the sticky bit on dir is set we should either
1880 * a. be owner of dir, or
1881 * b. be owner of victim, or
1882 * c. have CAP_FOWNER capability
1883 * 6. If the victim is append-only or immutable we can't do antyhing with
1884 * links pointing to it.
1885 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1886 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1887 * 9. We can't remove a root or mountpoint.
1888 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1889 * nfs_async_unlink().
1891 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1895 if (!victim->d_inode)
1898 BUG_ON(victim->d_parent->d_inode != dir);
1899 audit_inode_child(victim, dir);
1901 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1906 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1907 IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1910 if (!S_ISDIR(victim->d_inode->i_mode))
1912 if (IS_ROOT(victim))
1914 } else if (S_ISDIR(victim->d_inode->i_mode))
1916 if (IS_DEADDIR(dir))
1918 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1923 /* Check whether we can create an object with dentry child in directory
1925 * 1. We can't do it if child already exists (open has special treatment for
1926 * this case, but since we are inlined it's OK)
1927 * 2. We can't do it if dir is read-only (done in permission())
1928 * 3. We should have write and exec permissions on dir
1929 * 4. We can't do it if dir is immutable (done in permission())
1931 static inline int may_create(struct inode *dir, struct dentry *child)
1935 if (IS_DEADDIR(dir))
1937 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1941 * p1 and p2 should be directories on the same fs.
1943 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1948 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1952 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1954 p = d_ancestor(p2, p1);
1956 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1957 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1961 p = d_ancestor(p1, p2);
1963 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1964 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1968 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1969 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1973 void unlock_rename(struct dentry *p1, struct dentry *p2)
1975 mutex_unlock(&p1->d_inode->i_mutex);
1977 mutex_unlock(&p2->d_inode->i_mutex);
1978 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1982 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1983 struct nameidata *nd)
1985 int error = may_create(dir, dentry);
1990 if (!dir->i_op->create)
1991 return -EACCES; /* shouldn't it be ENOSYS? */
1994 error = security_inode_create(dir, dentry, mode);
1997 error = dir->i_op->create(dir, dentry, mode, nd);
1999 fsnotify_create(dir, dentry);
2003 static int may_open(struct path *path, int acc_mode, int flag)
2005 struct dentry *dentry = path->dentry;
2006 struct inode *inode = dentry->d_inode;
2016 switch (inode->i_mode & S_IFMT) {
2020 if (acc_mode & MAY_WRITE)
2025 if (path->mnt->mnt_flags & MNT_NODEV)
2034 error = inode_permission(inode, acc_mode);
2039 * An append-only file must be opened in append mode for writing.
2041 if (IS_APPEND(inode)) {
2042 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2048 /* O_NOATIME can only be set by the owner or superuser */
2049 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2055 static int handle_truncate(struct file *filp)
2057 struct path *path = &filp->f_path;
2058 struct inode *inode = path->dentry->d_inode;
2059 int error = get_write_access(inode);
2063 * Refuse to truncate files with mandatory locks held on them.
2065 error = locks_verify_locked(inode);
2067 error = security_path_truncate(path);
2069 error = do_truncate(path->dentry, 0,
2070 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2073 put_write_access(inode);
2077 static inline int open_to_namei_flags(int flag)
2079 if ((flag & O_ACCMODE) == 3)
2085 * Handle the last step of open()
2087 static struct file *do_last(struct nameidata *nd, struct path *path,
2088 const struct open_flags *op, const char *pathname)
2090 struct dentry *dir = nd->path.dentry;
2091 struct dentry *dentry;
2092 int open_flag = op->open_flag;
2093 int will_truncate = open_flag & O_TRUNC;
2095 int acc_mode = op->acc_mode;
2099 nd->flags &= ~LOOKUP_PARENT;
2100 nd->flags |= op->intent;
2102 switch (nd->last_type) {
2105 error = handle_dots(nd, nd->last_type);
2107 return ERR_PTR(error);
2110 error = complete_walk(nd);
2112 return ERR_PTR(error);
2113 audit_inode(pathname, nd->path.dentry);
2114 if (open_flag & O_CREAT) {
2120 error = complete_walk(nd);
2122 return ERR_PTR(error);
2123 audit_inode(pathname, dir);
2127 if (!(open_flag & O_CREAT)) {
2129 if (nd->last.name[nd->last.len])
2130 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2131 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2133 /* we _can_ be in RCU mode here */
2134 error = walk_component(nd, path, &nd->last, LAST_NORM,
2137 return ERR_PTR(error);
2138 if (error) /* symlink */
2141 error = complete_walk(nd);
2143 return ERR_PTR(error);
2146 if (nd->flags & LOOKUP_DIRECTORY) {
2147 if (!nd->inode->i_op->lookup)
2150 audit_inode(pathname, nd->path.dentry);
2154 /* create side of things */
2156 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2157 * cleared when we got to the last component we are about to look up
2159 error = complete_walk(nd);
2161 return ERR_PTR(error);
2163 audit_inode(pathname, dir);
2165 /* trailing slashes? */
2166 if (nd->last.name[nd->last.len])
2169 mutex_lock(&dir->d_inode->i_mutex);
2171 dentry = lookup_hash(nd);
2172 error = PTR_ERR(dentry);
2173 if (IS_ERR(dentry)) {
2174 mutex_unlock(&dir->d_inode->i_mutex);
2178 path->dentry = dentry;
2179 path->mnt = nd->path.mnt;
2181 /* Negative dentry, just create the file */
2182 if (!dentry->d_inode) {
2183 int mode = op->mode;
2184 if (!IS_POSIXACL(dir->d_inode))
2185 mode &= ~current_umask();
2187 * This write is needed to ensure that a
2188 * rw->ro transition does not occur between
2189 * the time when the file is created and when
2190 * a permanent write count is taken through
2191 * the 'struct file' in nameidata_to_filp().
2193 error = mnt_want_write(nd->path.mnt);
2195 goto exit_mutex_unlock;
2197 /* Don't check for write permission, don't truncate */
2198 open_flag &= ~O_TRUNC;
2200 acc_mode = MAY_OPEN;
2201 error = security_path_mknod(&nd->path, dentry, mode, 0);
2203 goto exit_mutex_unlock;
2204 error = vfs_create(dir->d_inode, dentry, mode, nd);
2206 goto exit_mutex_unlock;
2207 mutex_unlock(&dir->d_inode->i_mutex);
2208 dput(nd->path.dentry);
2209 nd->path.dentry = dentry;
2214 * It already exists.
2216 mutex_unlock(&dir->d_inode->i_mutex);
2217 audit_inode(pathname, path->dentry);
2220 if (open_flag & O_EXCL)
2223 error = follow_managed(path, nd->flags);
2228 nd->flags |= LOOKUP_JUMPED;
2231 if (!path->dentry->d_inode)
2234 if (path->dentry->d_inode->i_op->follow_link)
2237 path_to_nameidata(path, nd);
2238 nd->inode = path->dentry->d_inode;
2239 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
2240 error = complete_walk(nd);
2242 return ERR_PTR(error);
2244 if (S_ISDIR(nd->inode->i_mode))
2247 if (!S_ISREG(nd->inode->i_mode))
2250 if (will_truncate) {
2251 error = mnt_want_write(nd->path.mnt);
2257 error = may_open(&nd->path, acc_mode, open_flag);
2260 filp = nameidata_to_filp(nd);
2261 if (!IS_ERR(filp)) {
2262 error = ima_file_check(filp, op->acc_mode);
2265 filp = ERR_PTR(error);
2268 if (!IS_ERR(filp)) {
2269 if (will_truncate) {
2270 error = handle_truncate(filp);
2273 filp = ERR_PTR(error);
2279 mnt_drop_write(nd->path.mnt);
2280 path_put(&nd->path);
2284 mutex_unlock(&dir->d_inode->i_mutex);
2286 path_put_conditional(path, nd);
2288 filp = ERR_PTR(error);
2292 static struct file *path_openat(int dfd, const char *pathname,
2293 struct nameidata *nd, const struct open_flags *op, int flags)
2295 struct file *base = NULL;
2300 filp = get_empty_filp();
2302 return ERR_PTR(-ENFILE);
2304 filp->f_flags = op->open_flag;
2305 nd->intent.open.file = filp;
2306 nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2307 nd->intent.open.create_mode = op->mode;
2309 error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2310 if (unlikely(error))
2313 current->total_link_count = 0;
2314 error = link_path_walk(pathname, nd);
2315 if (unlikely(error))
2318 filp = do_last(nd, &path, op, pathname);
2319 while (unlikely(!filp)) { /* trailing symlink */
2320 struct path link = path;
2322 if (!(nd->flags & LOOKUP_FOLLOW)) {
2323 path_put_conditional(&path, nd);
2324 path_put(&nd->path);
2325 filp = ERR_PTR(-ELOOP);
2328 nd->flags |= LOOKUP_PARENT;
2329 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2330 error = follow_link(&link, nd, &cookie);
2331 if (unlikely(error))
2332 filp = ERR_PTR(error);
2334 filp = do_last(nd, &path, op, pathname);
2335 put_link(nd, &link, cookie);
2338 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2339 path_put(&nd->root);
2342 release_open_intent(nd);
2346 filp = ERR_PTR(error);
2350 struct file *do_filp_open(int dfd, const char *pathname,
2351 const struct open_flags *op, int flags)
2353 struct nameidata nd;
2356 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2357 if (unlikely(filp == ERR_PTR(-ECHILD)))
2358 filp = path_openat(dfd, pathname, &nd, op, flags);
2359 if (unlikely(filp == ERR_PTR(-ESTALE)))
2360 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2364 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2365 const char *name, const struct open_flags *op, int flags)
2367 struct nameidata nd;
2371 nd.root.dentry = dentry;
2373 flags |= LOOKUP_ROOT;
2375 if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2376 return ERR_PTR(-ELOOP);
2378 file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2379 if (unlikely(file == ERR_PTR(-ECHILD)))
2380 file = path_openat(-1, name, &nd, op, flags);
2381 if (unlikely(file == ERR_PTR(-ESTALE)))
2382 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2386 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2388 struct dentry *dentry = ERR_PTR(-EEXIST);
2389 struct nameidata nd;
2390 int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2392 return ERR_PTR(error);
2395 * Yucky last component or no last component at all?
2396 * (foo/., foo/.., /////)
2398 if (nd.last_type != LAST_NORM)
2400 nd.flags &= ~LOOKUP_PARENT;
2401 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2402 nd.intent.open.flags = O_EXCL;
2405 * Do the final lookup.
2407 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2408 dentry = lookup_hash(&nd);
2412 if (dentry->d_inode)
2415 * Special case - lookup gave negative, but... we had foo/bar/
2416 * From the vfs_mknod() POV we just have a negative dentry -
2417 * all is fine. Let's be bastards - you had / on the end, you've
2418 * been asking for (non-existent) directory. -ENOENT for you.
2420 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2422 dentry = ERR_PTR(-ENOENT);
2429 dentry = ERR_PTR(-EEXIST);
2431 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2436 EXPORT_SYMBOL(kern_path_create);
2438 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2440 char *tmp = getname(pathname);
2443 return ERR_CAST(tmp);
2444 res = kern_path_create(dfd, tmp, path, is_dir);
2448 EXPORT_SYMBOL(user_path_create);
2450 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2452 int error = may_create(dir, dentry);
2457 if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2458 !ns_capable(inode_userns(dir), CAP_MKNOD))
2461 if (!dir->i_op->mknod)
2464 error = devcgroup_inode_mknod(mode, dev);
2468 error = security_inode_mknod(dir, dentry, mode, dev);
2472 error = dir->i_op->mknod(dir, dentry, mode, dev);
2474 fsnotify_create(dir, dentry);
2478 static int may_mknod(mode_t mode)
2480 switch (mode & S_IFMT) {
2486 case 0: /* zero mode translates to S_IFREG */
2495 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2498 struct dentry *dentry;
2505 dentry = user_path_create(dfd, filename, &path, 0);
2507 return PTR_ERR(dentry);
2509 if (!IS_POSIXACL(path.dentry->d_inode))
2510 mode &= ~current_umask();
2511 error = may_mknod(mode);
2514 error = mnt_want_write(path.mnt);
2517 error = security_path_mknod(&path, dentry, mode, dev);
2519 goto out_drop_write;
2520 switch (mode & S_IFMT) {
2521 case 0: case S_IFREG:
2522 error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2524 case S_IFCHR: case S_IFBLK:
2525 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2526 new_decode_dev(dev));
2528 case S_IFIFO: case S_IFSOCK:
2529 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2533 mnt_drop_write(path.mnt);
2536 mutex_unlock(&path.dentry->d_inode->i_mutex);
2542 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2544 return sys_mknodat(AT_FDCWD, filename, mode, dev);
2547 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2549 int error = may_create(dir, dentry);
2554 if (!dir->i_op->mkdir)
2557 mode &= (S_IRWXUGO|S_ISVTX);
2558 error = security_inode_mkdir(dir, dentry, mode);
2562 error = dir->i_op->mkdir(dir, dentry, mode);
2564 fsnotify_mkdir(dir, dentry);
2568 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2570 struct dentry *dentry;
2574 dentry = user_path_create(dfd, pathname, &path, 1);
2576 return PTR_ERR(dentry);
2578 if (!IS_POSIXACL(path.dentry->d_inode))
2579 mode &= ~current_umask();
2580 error = mnt_want_write(path.mnt);
2583 error = security_path_mkdir(&path, dentry, mode);
2585 goto out_drop_write;
2586 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2588 mnt_drop_write(path.mnt);
2591 mutex_unlock(&path.dentry->d_inode->i_mutex);
2596 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2598 return sys_mkdirat(AT_FDCWD, pathname, mode);
2602 * The dentry_unhash() helper will try to drop the dentry early: we
2603 * should have a usage count of 2 if we're the only user of this
2604 * dentry, and if that is true (possibly after pruning the dcache),
2605 * then we drop the dentry now.
2607 * A low-level filesystem can, if it choses, legally
2610 * if (!d_unhashed(dentry))
2613 * if it cannot handle the case of removing a directory
2614 * that is still in use by something else..
2616 void dentry_unhash(struct dentry *dentry)
2618 shrink_dcache_parent(dentry);
2619 spin_lock(&dentry->d_lock);
2620 if (dentry->d_count == 1)
2622 spin_unlock(&dentry->d_lock);
2625 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2627 int error = may_delete(dir, dentry, 1);
2632 if (!dir->i_op->rmdir)
2636 mutex_lock(&dentry->d_inode->i_mutex);
2639 if (d_mountpoint(dentry))
2642 error = security_inode_rmdir(dir, dentry);
2646 shrink_dcache_parent(dentry);
2647 error = dir->i_op->rmdir(dir, dentry);
2651 dentry->d_inode->i_flags |= S_DEAD;
2655 mutex_unlock(&dentry->d_inode->i_mutex);
2662 static long do_rmdir(int dfd, const char __user *pathname)
2666 struct dentry *dentry;
2667 struct nameidata nd;
2669 error = user_path_parent(dfd, pathname, &nd, &name);
2673 switch(nd.last_type) {
2685 nd.flags &= ~LOOKUP_PARENT;
2687 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2688 dentry = lookup_hash(&nd);
2689 error = PTR_ERR(dentry);
2692 if (!dentry->d_inode) {
2696 error = mnt_want_write(nd.path.mnt);
2699 error = security_path_rmdir(&nd.path, dentry);
2702 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2704 mnt_drop_write(nd.path.mnt);
2708 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2715 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2717 return do_rmdir(AT_FDCWD, pathname);
2720 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2722 int error = may_delete(dir, dentry, 0);
2727 if (!dir->i_op->unlink)
2730 mutex_lock(&dentry->d_inode->i_mutex);
2731 if (d_mountpoint(dentry))
2734 error = security_inode_unlink(dir, dentry);
2736 error = dir->i_op->unlink(dir, dentry);
2741 mutex_unlock(&dentry->d_inode->i_mutex);
2743 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2744 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2745 fsnotify_link_count(dentry->d_inode);
2753 * Make sure that the actual truncation of the file will occur outside its
2754 * directory's i_mutex. Truncate can take a long time if there is a lot of
2755 * writeout happening, and we don't want to prevent access to the directory
2756 * while waiting on the I/O.
2758 static long do_unlinkat(int dfd, const char __user *pathname)
2762 struct dentry *dentry;
2763 struct nameidata nd;
2764 struct inode *inode = NULL;
2766 error = user_path_parent(dfd, pathname, &nd, &name);
2771 if (nd.last_type != LAST_NORM)
2774 nd.flags &= ~LOOKUP_PARENT;
2776 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2777 dentry = lookup_hash(&nd);
2778 error = PTR_ERR(dentry);
2779 if (!IS_ERR(dentry)) {
2780 /* Why not before? Because we want correct error value */
2781 if (nd.last.name[nd.last.len])
2783 inode = dentry->d_inode;
2787 error = mnt_want_write(nd.path.mnt);
2790 error = security_path_unlink(&nd.path, dentry);
2793 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2795 mnt_drop_write(nd.path.mnt);
2799 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2801 iput(inode); /* truncate the inode here */
2808 error = !dentry->d_inode ? -ENOENT :
2809 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2813 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2815 if ((flag & ~AT_REMOVEDIR) != 0)
2818 if (flag & AT_REMOVEDIR)
2819 return do_rmdir(dfd, pathname);
2821 return do_unlinkat(dfd, pathname);
2824 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2826 return do_unlinkat(AT_FDCWD, pathname);
2829 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2831 int error = may_create(dir, dentry);
2836 if (!dir->i_op->symlink)
2839 error = security_inode_symlink(dir, dentry, oldname);
2843 error = dir->i_op->symlink(dir, dentry, oldname);
2845 fsnotify_create(dir, dentry);
2849 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2850 int, newdfd, const char __user *, newname)
2854 struct dentry *dentry;
2857 from = getname(oldname);
2859 return PTR_ERR(from);
2861 dentry = user_path_create(newdfd, newname, &path, 0);
2862 error = PTR_ERR(dentry);
2866 error = mnt_want_write(path.mnt);
2869 error = security_path_symlink(&path, dentry, from);
2871 goto out_drop_write;
2872 error = vfs_symlink(path.dentry->d_inode, dentry, from);
2874 mnt_drop_write(path.mnt);
2877 mutex_unlock(&path.dentry->d_inode->i_mutex);
2884 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2886 return sys_symlinkat(oldname, AT_FDCWD, newname);
2889 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2891 struct inode *inode = old_dentry->d_inode;
2897 error = may_create(dir, new_dentry);
2901 if (dir->i_sb != inode->i_sb)
2905 * A link to an append-only or immutable file cannot be created.
2907 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2909 if (!dir->i_op->link)
2911 if (S_ISDIR(inode->i_mode))
2914 error = security_inode_link(old_dentry, dir, new_dentry);
2918 mutex_lock(&inode->i_mutex);
2919 /* Make sure we don't allow creating hardlink to an unlinked file */
2920 if (inode->i_nlink == 0)
2923 error = dir->i_op->link(old_dentry, dir, new_dentry);
2924 mutex_unlock(&inode->i_mutex);
2926 fsnotify_link(dir, inode, new_dentry);
2931 * Hardlinks are often used in delicate situations. We avoid
2932 * security-related surprises by not following symlinks on the
2935 * We don't follow them on the oldname either to be compatible
2936 * with linux 2.0, and to avoid hard-linking to directories
2937 * and other special files. --ADM
2939 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2940 int, newdfd, const char __user *, newname, int, flags)
2942 struct dentry *new_dentry;
2943 struct path old_path, new_path;
2947 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2950 * To use null names we require CAP_DAC_READ_SEARCH
2951 * This ensures that not everyone will be able to create
2952 * handlink using the passed filedescriptor.
2954 if (flags & AT_EMPTY_PATH) {
2955 if (!capable(CAP_DAC_READ_SEARCH))
2960 if (flags & AT_SYMLINK_FOLLOW)
2961 how |= LOOKUP_FOLLOW;
2963 error = user_path_at(olddfd, oldname, how, &old_path);
2967 new_dentry = user_path_create(newdfd, newname, &new_path, 0);
2968 error = PTR_ERR(new_dentry);
2969 if (IS_ERR(new_dentry))
2973 if (old_path.mnt != new_path.mnt)
2975 error = mnt_want_write(new_path.mnt);
2978 error = security_path_link(old_path.dentry, &new_path, new_dentry);
2980 goto out_drop_write;
2981 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
2983 mnt_drop_write(new_path.mnt);
2986 mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2987 path_put(&new_path);
2989 path_put(&old_path);
2994 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2996 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3000 * The worst of all namespace operations - renaming directory. "Perverted"
3001 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3003 * a) we can get into loop creation. Check is done in is_subdir().
3004 * b) race potential - two innocent renames can create a loop together.
3005 * That's where 4.4 screws up. Current fix: serialization on
3006 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3008 * c) we have to lock _three_ objects - parents and victim (if it exists).
3009 * And that - after we got ->i_mutex on parents (until then we don't know
3010 * whether the target exists). Solution: try to be smart with locking
3011 * order for inodes. We rely on the fact that tree topology may change
3012 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3013 * move will be locked. Thus we can rank directories by the tree
3014 * (ancestors first) and rank all non-directories after them.
3015 * That works since everybody except rename does "lock parent, lookup,
3016 * lock child" and rename is under ->s_vfs_rename_mutex.
3017 * HOWEVER, it relies on the assumption that any object with ->lookup()
3018 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3019 * we'd better make sure that there's no link(2) for them.
3020 * d) conversion from fhandle to dentry may come in the wrong moment - when
3021 * we are removing the target. Solution: we will have to grab ->i_mutex
3022 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3023 * ->i_mutex on parents, which works but leads to some truly excessive
3026 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3027 struct inode *new_dir, struct dentry *new_dentry)
3030 struct inode *target = new_dentry->d_inode;
3033 * If we are going to change the parent - check write permissions,
3034 * we'll need to flip '..'.
3036 if (new_dir != old_dir) {
3037 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3042 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3048 mutex_lock(&target->i_mutex);
3051 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3055 shrink_dcache_parent(new_dentry);
3056 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3061 target->i_flags |= S_DEAD;
3062 dont_mount(new_dentry);
3066 mutex_unlock(&target->i_mutex);
3069 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3070 d_move(old_dentry,new_dentry);
3074 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3075 struct inode *new_dir, struct dentry *new_dentry)
3077 struct inode *target = new_dentry->d_inode;
3080 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3086 mutex_lock(&target->i_mutex);
3089 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3092 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3097 dont_mount(new_dentry);
3098 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3099 d_move(old_dentry, new_dentry);
3102 mutex_unlock(&target->i_mutex);
3107 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3108 struct inode *new_dir, struct dentry *new_dentry)
3111 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3112 const unsigned char *old_name;
3114 if (old_dentry->d_inode == new_dentry->d_inode)
3117 error = may_delete(old_dir, old_dentry, is_dir);
3121 if (!new_dentry->d_inode)
3122 error = may_create(new_dir, new_dentry);
3124 error = may_delete(new_dir, new_dentry, is_dir);
3128 if (!old_dir->i_op->rename)
3131 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3134 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3136 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3138 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3139 new_dentry->d_inode, old_dentry);
3140 fsnotify_oldname_free(old_name);
3145 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3146 int, newdfd, const char __user *, newname)
3148 struct dentry *old_dir, *new_dir;
3149 struct dentry *old_dentry, *new_dentry;
3150 struct dentry *trap;
3151 struct nameidata oldnd, newnd;
3156 error = user_path_parent(olddfd, oldname, &oldnd, &from);
3160 error = user_path_parent(newdfd, newname, &newnd, &to);
3165 if (oldnd.path.mnt != newnd.path.mnt)
3168 old_dir = oldnd.path.dentry;
3170 if (oldnd.last_type != LAST_NORM)
3173 new_dir = newnd.path.dentry;
3174 if (newnd.last_type != LAST_NORM)
3177 oldnd.flags &= ~LOOKUP_PARENT;
3178 newnd.flags &= ~LOOKUP_PARENT;
3179 newnd.flags |= LOOKUP_RENAME_TARGET;
3181 trap = lock_rename(new_dir, old_dir);
3183 old_dentry = lookup_hash(&oldnd);
3184 error = PTR_ERR(old_dentry);
3185 if (IS_ERR(old_dentry))
3187 /* source must exist */
3189 if (!old_dentry->d_inode)
3191 /* unless the source is a directory trailing slashes give -ENOTDIR */
3192 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3194 if (oldnd.last.name[oldnd.last.len])
3196 if (newnd.last.name[newnd.last.len])
3199 /* source should not be ancestor of target */
3201 if (old_dentry == trap)
3203 new_dentry = lookup_hash(&newnd);
3204 error = PTR_ERR(new_dentry);
3205 if (IS_ERR(new_dentry))
3207 /* target should not be an ancestor of source */
3209 if (new_dentry == trap)
3212 error = mnt_want_write(oldnd.path.mnt);
3215 error = security_path_rename(&oldnd.path, old_dentry,
3216 &newnd.path, new_dentry);
3219 error = vfs_rename(old_dir->d_inode, old_dentry,
3220 new_dir->d_inode, new_dentry);
3222 mnt_drop_write(oldnd.path.mnt);
3228 unlock_rename(new_dir, old_dir);
3230 path_put(&newnd.path);
3233 path_put(&oldnd.path);
3239 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3241 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3244 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3248 len = PTR_ERR(link);
3253 if (len > (unsigned) buflen)
3255 if (copy_to_user(buffer, link, len))
3262 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3263 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3264 * using) it for any given inode is up to filesystem.
3266 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3268 struct nameidata nd;
3273 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3275 return PTR_ERR(cookie);
3277 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3278 if (dentry->d_inode->i_op->put_link)
3279 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3283 int vfs_follow_link(struct nameidata *nd, const char *link)
3285 return __vfs_follow_link(nd, link);
3288 /* get the link contents into pagecache */
3289 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3293 struct address_space *mapping = dentry->d_inode->i_mapping;
3294 page = read_mapping_page(mapping, 0, NULL);
3299 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3303 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3305 struct page *page = NULL;
3306 char *s = page_getlink(dentry, &page);
3307 int res = vfs_readlink(dentry,buffer,buflen,s);
3310 page_cache_release(page);
3315 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3317 struct page *page = NULL;
3318 nd_set_link(nd, page_getlink(dentry, &page));
3322 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3324 struct page *page = cookie;
3328 page_cache_release(page);
3333 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3335 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3337 struct address_space *mapping = inode->i_mapping;
3342 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3344 flags |= AOP_FLAG_NOFS;
3347 err = pagecache_write_begin(NULL, mapping, 0, len-1,
3348 flags, &page, &fsdata);
3352 kaddr = kmap_atomic(page, KM_USER0);
3353 memcpy(kaddr, symname, len-1);
3354 kunmap_atomic(kaddr, KM_USER0);
3356 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3363 mark_inode_dirty(inode);
3369 int page_symlink(struct inode *inode, const char *symname, int len)
3371 return __page_symlink(inode, symname, len,
3372 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3375 const struct inode_operations page_symlink_inode_operations = {
3376 .readlink = generic_readlink,
3377 .follow_link = page_follow_link_light,
3378 .put_link = page_put_link,
3381 EXPORT_SYMBOL(user_path_at);
3382 EXPORT_SYMBOL(follow_down_one);
3383 EXPORT_SYMBOL(follow_down);
3384 EXPORT_SYMBOL(follow_up);
3385 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3386 EXPORT_SYMBOL(getname);
3387 EXPORT_SYMBOL(lock_rename);
3388 EXPORT_SYMBOL(lookup_one_len);
3389 EXPORT_SYMBOL(page_follow_link_light);
3390 EXPORT_SYMBOL(page_put_link);
3391 EXPORT_SYMBOL(page_readlink);
3392 EXPORT_SYMBOL(__page_symlink);
3393 EXPORT_SYMBOL(page_symlink);
3394 EXPORT_SYMBOL(page_symlink_inode_operations);
3395 EXPORT_SYMBOL(kern_path);
3396 EXPORT_SYMBOL(vfs_path_lookup);
3397 EXPORT_SYMBOL(inode_permission);
3398 EXPORT_SYMBOL(unlock_rename);
3399 EXPORT_SYMBOL(vfs_create);
3400 EXPORT_SYMBOL(vfs_follow_link);
3401 EXPORT_SYMBOL(vfs_link);
3402 EXPORT_SYMBOL(vfs_mkdir);
3403 EXPORT_SYMBOL(vfs_mknod);
3404 EXPORT_SYMBOL(generic_permission);
3405 EXPORT_SYMBOL(vfs_readlink);
3406 EXPORT_SYMBOL(vfs_rename);
3407 EXPORT_SYMBOL(vfs_rmdir);
3408 EXPORT_SYMBOL(vfs_symlink);
3409 EXPORT_SYMBOL(vfs_unlink);
3410 EXPORT_SYMBOL(dentry_unhash);
3411 EXPORT_SYMBOL(generic_readlink);